Parallel computation of GS algorithm and its robustness to rounding-off error and noise

Abstract

We use a graphics processing unit (GPU) to speed up the calculation of the  Fourier transform in the Gerchberg Saxton (GS) algorithm, angular spectrum method and the correlation integral. The use of the cuFFT GPU implementation of the Fourier transformation process decreases significantly the average runtime of the aforementioned methods which are useful tools in information and signal processing. We examine the robustness of GS algorithm against the presence of rounding-off errors and noise in the target image and found that reconstruction fidelity deteriorates nonlinearly with decrease in the bit number representation or signal-to-noise ratio of the target image.